On the mean value of the product of multiplicative functions with shifted argument (Q2385070): Difference between revisions

The authors are interested in the asymptotic behaviour of the mean-value function \[ M_x(g,f) = \frac1x \cdot \sum_{n\leq x} g(n) \cdot f(n+1), \text{ as } x\to\infty, \] for multiplicative functions \(f\), \(g\) of absolute value \(|f|\leq 1\), \(|g|\leq 1\). Abbreviating \[ \eta(n) = \eta_\tau(n) = \prod_{p\mid n}\left(1+ \sum_{k=1}^\infty \frac{f(p^k)}{p^{k(1+i\tau)}} \right)^{-1}, \text{ where }\tau\in\mathbb R, \] \[ P(x) = \prod_{\log x < p\leq x} \left(1-\frac1p\right) \left( 1 + \sum_{k=1}^\infty \frac{g(p^k)}{p^k} \right), \] and \[ \Pi(x) = \Pi_\tau(x) = \prod_{p\leq \log x} \left( 1 - \frac{\eta_\tau(p)}{p-1} + \eta_\tau(p) \cdot \sum_{k=1}^\infty \frac{g(p^k)}{p^k} \right), \] the authors' main result is the asymptotic formula [holding uniformly for \(x\geq 2\)] \[ M_x(g,f) = \left(\frac1x \sum_{n\leq x} f(n)\right) \cdot \Pi_\tau(x) \cdot P(x) + {\mathcal O}\left( \sqrt{\sum_{\log x < p \leq x}\frac{|g(p) -1|^2}p} + (\log x)^{-c} \right) \] for some constant \(c>0\) and some \(\tau \in \mathbb R\), \(|\tau| \leq (\log x)^{1/19}\), \textbf{if} \(f\) is [nearly] constant on the primes in the mean, i.e. there is some \(\varkappa \in \mathbb C\) such that \[ \sum_{p\leq x} |f(p) - \varkappa | \ll_t \frac x{\log^t(x)} \text{ for any } t>0, \] and if there is some \(c_1>0\) so that \[ \left|1 + \sum_{k=1}^\infty \frac{f(2^k)}{2^{k(1+i\xi)}}\right| \geq c_1 \qquad \begin{cases} \text{for } \xi=0, &\text{if \(f\) is real-valued},\\ \text{for all } \xi, &\text{otherwise}. \end{cases} \] For real-valued functions \(f\) one may take \(\tau=0\). In particular the result is applicable to the Liouville function \(\lambda\), \(\lambda(n) = (-1)^{\Omega(n)}\), \[ \frac1x \cdot \sum_{n\leq x} g(n) \cdot\lambda(n+1) = {\mathcal O}\left( \sqrt{\sum_{\log x < p \leq x}\frac{|g(p) -1|^2}p} + (\log x)^{-c} \right), \] and to sums \[ \frac1x \cdot \sum_{n\leq x} g(n)\cdot \varepsilon^{\omega(n+1)} \] with a complex root of unity \(\varepsilon \not= \pm 1\) and to \[ \frac1x \sum_{n\leq x, \; (n+1,\; m)=1} g(n). \] The proof uses the Turán-Kubilius inequality, and deep results from the theory of multiplicative functions concerning \(\sum_{n\leq x,\;(n,\; d)=1} f(n)\) due to \textit{P. D. T. A. Elliott} [Indag. Math. 51, No. 4, 409--420 (1989; Zbl 0696.10041)], the well-known mean-value theorem of G. Halász (1968, 1971) in a version with sharper remainder term due too \textit{G. Stepanauskas} [Lith. Math. J. 27, No. 2, 159--163 (1987; Zbl 0643.10036)] and a Bombieri-Vinogradov result for multiplicative functions due to \textit{D. Wolke} [Math. Ann. 202, 1--25 (1973; Zbl 0238.10031)].